MCP Tools

Generate valid OTP/OJP documents from any input — resumes, job posting PDFs, plain English descriptions — using Claude and the ADNX MCP server. No manual JSON editing required.

The MCP server exposes 6 tools that handle parsing, validation, and introspection for both protocols. Claude orchestrates the full pipeline: extract structured data from unstructured input, fill the schema skeleton, validate against OTP/OJP v0.2, and optionally submit directly to the sandbox API.

---

Installation

Claude Code

Add the MCP server to your project:

Add MCP server

claude mcp add open-talent-protocol -- npx @opentalentprotocol/mcp-server

Claude Desktop

Add to your claude_desktop_config.json:

claude_desktop_config.json

{
  "mcpServers": {
    "open-talent-protocol": {
      "command": "npx",
      "args": ["@opentalentprotocol/mcp-server"]
    }
  }
}

Requirements

Node.js 18+ and npm. The MCP server is MIT-licensed and runs locally — no API keys, no data leaves your machine.

---

Workflow

The tools are designed as a pipeline. Claude calls them in sequence:

Parse → Validate → Submit

1. PARSE     Paste resume/job posting text into Claude
              → otp_parse_resume / ojp_parse_job_posting
              → Returns skeleton with _EXTRACT_* annotations + field confidence

2. FILL       Claude fills the skeleton using its reasoning over the source text
              → Resolves annotations, normalizes dates/currencies/languages

3. VALIDATE   → otp_validate_profile / ojp_validate_job_posting
              → Returns errors (if any) with exact field paths

4. FIX        Claude fixes validation errors and re-validates

5. USE        Copy the JSON, download it, or submit directly to the sandbox API
              → POST /api/v1/talent (OTP) or POST /api/v1/jobs (OJP)

One-shot prompts

You don’t need to call tools manually. Just tell Claude what you want:

• “Parse this resume into OTP v0.2” + paste the text
• “Convert this job posting to OJP and submit it to the sandbox”
• “I need a Senior Backend Engineer in Berlin, 90-110k EUR, must know Go and PostgreSQL”

---

Tools

Three tools per protocol: parse (unstructured → structured), validate (check against schema), introspect (summarize for agents).

otp_parse_resume

Produces an OTP extraction template from raw resume text. Returns a document skeleton with _EXTRACT_* annotations that Claude fills in, plus confidence levels and a list of gaps (info not typically found on resumes).

Property	Type	Required	Description
text	string	yes	Raw resume text (plain text, markdown, or extracted PDF text)

Returns

{
  "skeleton": {
    "schema_version": "0.2.0",
    "otp_id": "_EXTRACT_UUID",
    "disclosure_tier": "profile",
    "name": { "given": "_EXTRACT_GIVEN_NAME", "family": "_EXTRACT_FAMILY_NAME" },
    "title": "_EXTRACT_CURRENT_TITLE",
    "skills": "_EXTRACT_SKILLS_ARRAY"
  },
  "fieldConfidence": [
    { "field": "name", "confidence": "high" },
    { "field": "salary_band", "confidence": "low" }
  ],
  "gaps": [
    "salary_band — rarely on resumes, ask candidate",
    "availability — not typically stated",
    "consent_date — must be provided by submitting agent"
  ]
}

otp_validate_profile

Validates a JSON document against the OTP v0.2 schema. Returns validity status and field-level errors.

Property	Type	Required	Description
document	object	no	A pre-parsed OTP document object
filePath	string	no	Absolute path to an OTP JSON file

Provide either document or filePath.

Returns (valid)

{ "valid": true, "errors": [] }

Returns (invalid)

{
  "valid": false,
  "errors": [
    { "path": "/skills/0/level", "message": "must be integer between 1 and 5" },
    { "path": "/location", "message": "must have required property 'country'" }
  ]
}

otp_introspect_profile

Extracts a structured, agent-friendly summary from an OTP document. Returns normalized skills, work history, preferences, and an agentSummary string suitable for use in a system prompt.

Property	Type	Required	Description
document	object	no	A pre-parsed OTP document object
filePath	string	no	Absolute path to an OTP JSON file
sections	string[]	no	Filter: identity, summary, skills, work, preferences (default: all)

ojp_parse_job_posting

Produces an OJP extraction template from raw job posting text. Returns a document skeleton with _EXTRACT_* annotations, field confidence levels, and a gaps list of information commonly missing from postings.

Property	Type	Required	Description
text	string	yes	Raw job posting text (plain text, markdown, or extracted HTML)
sourceUrl	string	no	Source URL of the job posting, if known

Returns

{
  "skeleton": {
    "schema_version": "0.2.0",
    "ojp_id": "_EXTRACT_UUID",
    "status": "active",
    "title": "_EXTRACT_JOB_TITLE",
    "must_have": {
      "skills": "_EXTRACT_REQUIRED_SKILLS",
      "languages": "_EXTRACT_REQUIRED_LANGUAGES"
    }
  },
  "fieldConfidence": [
    { "field": "title", "confidence": "high" },
    { "field": "salary_band", "confidence": "medium" }
  ],
  "gaps": [
    "salary_band — not always disclosed in postings",
    "organization.size — rarely stated explicitly",
    "must_have.experience_years — often vague ('experienced')"
  ]
}

ojp_validate_job_posting

Validates a JSON document against the OJP v0.2 schema. Returns validity status and field-level errors.

Property	Type	Required	Description
document	object	no	A pre-parsed OJP document object
filePath	string	no	Absolute path to an OJP JSON file

Provide either document or filePath.

Returns (valid)

{ "valid": true, "errors": [] }

Returns (invalid)

{
  "valid": false,
  "errors": [
    { "path": "/must_have/skills/0", "message": "must have required property 'name'" },
    { "path": "/employment_type", "message": "must be one of: full_time, part_time, contract, freelance, internship, temporary" }
  ]
}

ojp_introspect_job_posting

Extracts a structured, agent-friendly summary from an OJP document. Returns normalized requirements, compensation, location, team, process details, and an agentSummary string suitable for use in a system prompt.

Property	Type	Required	Description
document	object	no	A pre-parsed OJP document object
filePath	string	no	Absolute path to an OJP JSON file

---

Examples

Copy-paste these prompts into Claude (with the MCP server installed) to generate valid documents.

Resume → OTP

Paste a resume (text or PDF) and ask Claude to convert it:

Prompt

Parse this resume into an OTP v0.2 profile, validate it, and show me the JSON:

Lena Muller
Senior Backend Engineer | Berlin, Germany

Experience:
- Senior Backend Engineer at Delivery Hero SE (Mar 2022 - present)
  Built high-throughput order processing pipeline in Go and Rust.

- Backend Engineer at Zalando SE (Jun 2018 - Feb 2022)
  PostgreSQL optimization, API gateway development.

Skills: Rust (5 years), Go (6 years), PostgreSQL (8 years)
Languages: German (native), English (C1)
Looking for: Remote/hybrid, 80-100k EUR, available in 2 weeks

Claude will call otp_parse_resume → fill the skeleton → otp_validate_profile → return valid OTP v0.2 JSON.

Job posting → OJP

Describe a role in plain English or paste a job posting:

Prompt

Convert this to OJP v0.2 and validate:

We're hiring a Senior Backend Engineer at Acme Corp (scale-up, Berlin).
Hybrid, 85-110k EUR.

Must have:
- Go (4+ years, advanced)
- PostgreSQL (intermediate+)
- 5+ years experience
- English C1

Nice to have: Rust

Need work authorization for DE/AT/CH.

Claude will call ojp_parse_job_posting → fill the skeleton → ojp_validate_job_posting → return valid OJP v0.2 JSON.

Generate → Submit to sandbox

Combine MCP tools with the sandbox API in a single prompt:

Prompt

I have a sandbox API key: adnx_test_k1_a3f8e2b1c4d5

1. Parse the job posting below into OJP v0.2
2. Validate it
3. Submit it to https://sandbox.adnx.ai/api/v1/jobs

[paste job posting text here]

End-to-end automation

With Claude Code, you can script the full pipeline: parse unstructured input → validate → submit to the sandbox → check the negotiation result. The MCP tools handle schema compliance; the sandbox API handles matching.